report from dagstuhl seminar 16062 modeling and analysis ...€¦ · report from dagstuhl seminar...

Report from Dagstuhl Seminar 16062

Modeling and Analysis of Semiconductor Supply ChainsEdited byChen-Fu Chien1, Hans Ehm2, John Fowler3, and Lars Mönch4

1 Department of Industrial Engineering & Engineering Management, NationalTsing Hua University, Hsinchu, Taiwan, [email protected]

2 Supply Chain Innovation, Infineon Technologies AG, München, Germany,[email protected]

3 Department of Supply Chain Management, Arizona State University, Tempe,USA, [email protected]

4 Chair of Enterprise-wide Software Systems, University of Hagen, Germany,[email protected]

AbstractIn February 2016 the Dagstuhl Seminar 16062 explored the needs of the semiconductor industryfor better planning and scheduling approaches at the supply chain level and the requirementsfor information systems to support the approaches. The seminar participants also spent timeidentifying the core elements of a conceptual reference model for planning and control of semi-conductor manufacturing supply chains. This Executive Summary describes the process of theseminar and discusses key findings and areas for future research regarding these topics. Abstractsof presentations given during the seminar and the output of breakout sessions are collected inappendices.

Seminar February 7–12, 2016 – http://www.dagstuhl.de/160621998 ACM Subject Classification B.2.2 Performance Analysis and Design Aids (Simulation),

F.2.2 Nonnumerical Algorithms and Problems (Scheduling and Sequencing), H.4.2 Types ofSystems (Decision Support, Logistics), I.2.8 Problem Solving, Control Methods, and Search(Heuristic methods, Plan execution, formation, and generation, Scheduling)

Keywords and phrases Modeling, Simulation, Supply Chain Management, Semiconductor Man-ufacturing

Digital Object Identifier 10.4230/DagRep.6.2.28Edited in cooperation with Raphael Herding

1 Summary

Chen-Fu ChienHans EhmLars Mönch

License Creative Commons BY 3.0 Unported license© Chen-Fu Chien, Hans Ehm, and Lars Mönch

Complex manufacturing processes are the heart of semiconductor manufacturing. A semi-conductor chip is a highly miniaturized, integrated circuit (IC) consisting of thousands ofcomponents. Semiconductor manufacturing starts with thin discs, called wafers, (typically)made of silicon. A large number of usually identical chips can be produced on each wafer byfabricating the ICs layer by layer in a wafer fabrication facility (wafer fab). The correspondingstep is referred to as the Fab step. Next, electrical tests that identify the individual dies that

Except where otherwise noted, content of this report is licensedunder a Creative Commons BY 3.0 Unported license

Modeling and Analysis of Semiconductor Supply Chains, Dagstuhl Reports, Vol. 6, Issue 2, pp. 28–64Editors: Chen-Fu Chien, Hans Ehm, John Fowler, and Lars Mönch

Dagstuhl ReportsSchloss Dagstuhl – Leibniz-Zentrum für Informatik, Dagstuhl Publishing, Germany

http://www.dagstuhl.de/16062

http://dx.doi.org/10.4230/DagRep.6.2.28

http://creativecommons.org/licenses/by/3.0/




http://www.dagstuhl.de/dagstuhl-reports/

http://www.dagstuhl.de

Chen-Fu Chien, Hans Ehm, John Fowler, and Lars Mönch 29

are likely to fail when packaged are performed in the Probe facility. An electronic map of thecondition of each die is made so that only the good ones will be used. The probed wafers arethen sent to an Assembly facility where the good dies are put into an appropriate package.The assembled dies are sent to a test facility where they are tested to ensure that only goodproducts are sent to customers. The tested devices are then sent to regional warehousesor directly to customers. Wafer fabrication and probe are often called the front-end andassembly and test are called the back-end.

Supply chain management (SCM) problems have become more and more important in thelast decade. This has been caused by the fact that front-end operations are often performed inhighly industrialized nations, while back-end operations are typically carried out in countrieswhere labor rates are cheaper. Moreover, there are centers of competencies (e.g. bumping)that may consist of only a few process steps that may be done in a different company ownedfacility or remotely by a subcontractor. These centers of competencies speed up innovationsand reduce costs, but increase the complexity of SCM.

The semiconductor industry is capital intensive with the cost of an entire wafer fab up tonearly $10 billion US caused primarily by extremely expensive machines, some up to $100million US each. The manufacturing process is very complex due to the reentrant flows incombination with very long cycle times and the multiple sources of uncertainty involved.Capacity expansions are very expensive and time-consuming. This kind of decision is basedon demand forecasts for the next years. Because of the rapidly changing environment, thedemand is highly volatile. Consequently, the forecast is rarely accurate. The semiconductorindustry is an extreme field for SCM solutions from an algorithmic as well as from a softwareand information systems point of view. The huge size of the supply chains involved, thepervasive presence of different kinds of uncertainties and the rapid pace of change leads toan environment that places approaches developed in other industries under major stress.Modeling and analysis approaches that are successful in this industry are likely to findapplications in other areas, and to significantly advance the state of the art in their fields(cf. [1]).

The purpose of this seminar was to bring together researchers from different disciplinesincluding information systems, computer science, industrial engineering, operations research,and supply chain management whose central interest is in modeling, analyzing, and designingcomplex and large-scale supply chains as in the semiconductor industry. Moreover, practi-tioners from the semiconductor industry who have frequently articulated their perceptionthat academic research does not always address the real problems faced by the industrybrought in their domain knowledge to make sure that progress towards applicability andfeasibility would be made during this seminar. The seminar had 26 attendees from tendifferent countries (see participant list at the end of the report). We had participants fromleading semiconductor companies Infineon Technologies and Intel Corp. as well as researcherswho work closely with ST Microelectronics, Globalfoundries, and Taiwan SemiconductorManufacturing Company (TSMC).

A primary purpose of the workshop was to extend the scope of the academic researchcommunity from single wafer fabs to the entire semiconductor supply chain. We show theprinciple architecture of the planning and control system of a semiconductor supply chain inFigure 1.

16062

30 16062 – Modeling and Analysis of Semiconductor Supply Chains

Figure 1 Planning and Control System of a Semiconductor Supply Chain (adapted from [2]).

Seminar ObjectivesThe first objective of the seminar consisted of developing a research agenda for semiconductorsupply chain modeling and analysis topics. This includes innovative modeling approaches forsupply chain network planning, demand planning, master planning, and detailed productionplanning and scheduling in semiconductor supply chains. But it also includes ideas on howto design the related future information systems.

The research agenda was developed around the following two main topics:

Topic 1: Novel planning and scheduling approaches that can deal with the complexityand stochasticity of the semiconductor supply chain:

Many planning approaches on the SC-level are based on (distributed) hierarchicaland generally deterministic approaches to deal with the sheer complexity of thesemiconductor supply chain. The role of anticipation of lower level behavior in upperlevel decision-making is still not well understood and has to be studied in more detail.Because a semiconductor supply chain contains many different, often autonomousdecision-making entities including humans, negotiation approaches are typical in suchdistributed hierarchical systems for planning and control. It should be researched howsuch negotiation approaches can be automated and which decisions should be made byhumans.The overall cycle times in a typical semiconductor supply chain are on the order of 10to 15 weeks. Therefore lead times have to be modeled in planning formulations. Usinglead times as exogenous parameters in planning formulations leads to a well-knowncircularity because the cycle time depends in a nonlinear manner on the resourceutilization which is a result of the release decisions made by the planning approach.


Different types of clearing functions have to be researched in the semiconductor supplychain context.Approaches to demand planning that take the product life cycle into account have tobe studied. The interaction of demand planning and supply chain planning has to beinvestigated.Different ways to anticipate stochasticity including robust optimization, approximatedynamic programming, and stochastic programming have to be researched in thesemiconductor supply chain context.Different ways to appropriately deal with stochasticity including rolling planningtechniques and inventory holding strategies have to be studied.Generation of scenarios and other distribution parameters for planning problems insupply chains using data mining techniques have to be researched.Because of the complexity of supply chains, long computing times still hinder the usageof analytic solution approaches especially for what-if analysis. The role of state-of-the-art computing techniques including parallel computing on Graphics Processing Units(GPU) machines or Cloud computing techniques in decision-making for semiconductorsupply chains has to be investigated.

Topic 2: Future information systems and supply chain management in the semicon-ductor industry:

Understanding the limitations of today’s packaged software for supply chain manage-ment in the semiconductor industry.Proposing alternative software solutions including software agents and service-orientedcomputing for planning and scheduling applications in the supply chain context.Integration concepts for state-of-the-art computing techniques to get models that arecomputationally tractable and address the different uncertainties encountered in thisindustry.Approaches to embed real time simulation techniques in current and future informationsystems to support decision-making in semiconductor supply chains.Understanding the interaction of human agents with information systems.

The implementation of ERP, APS, and MES systems in semiconductor supply chains providesboth an opportunity and the need for development of supply-chain wide integrated productionplanning and scheduling solutions. Therefore, we think that the second topic is importantand should be also addressed in the research agenda. Research related only to the first maintopic is not sufficient.

The second objective of the seminar consisted of identifying the core elements of aconceptual reference model for planning and control of a supply chain in the semiconductorindustry that can be used for analysis and performance assessment purposes and to fostera common understanding in the research community both in academia and industry. Thisincluded specifying reference planning and control activities, the major information flows,and their interaction with a reference system of a physical supply chain. Due to the inherentcomplexity of semiconductor supply chains it requires simulation of the physical supplychain to understand the interactions between the planning and control components andthe physical supply chain, to find solution approaches to problems and to verify them inthe risk-free simulation environment before implementing them. There are widely acceptedreference (simulation) models for single wafer fabs, mainly developed in the Measurementand Improvement of Manufacturing Capacity (MIMAC) project (led by one of the organizersof this Dagstuhl seminar) 20 years ago that are still used by many academic researchersworking with the semiconductor industry.

16062


Existing reference models on the planning and control level like the Supply ChainOperations (SCOR) reference model and the supply chain planning (SCP) matrix are toogeneric to be useful for detailed analysis and have to be refined considerably to cover theimportant domain-specific aspects of semiconductor supply chains.

The ProcessIn the opening session, the organizers welcomed the participants and acknowledged InfineonTechnologies as a sponsor of the seminar. Next, the participants each introduced themselves.This was followed by an overview of the goals and objectives of the seminar and a detailedreview of the seminar program including the ground rules for interactions.

The remainder of the day on Monday consisted of four industry overview talks (by HansEhm, Kenneth Fordyce, Chen-Fu Chien, and Irfan Ovacik) and a review of the literaturerelated to modeling an analysis of semiconductor supply chains (by Lars Mönch and RehaUzsoy). Tuesday and half a day on Wednesday were devoted to presentations and discussionsabout the various elements of the semiconductor supply chain planning and control systemsshown in Figure 1 above. See Table 1 for a list of topics and presenters and Section 3 forabstracts of the presentations.

Wednesday afternoon was the excursion that was enjoyed by the participants. Thursdaywas devoted to 3 breakout sessions with report outs on the topics in Table 2. Section 4 hasthe breakout report outs.

The first set of breakout sessions had four groups focus on the individual elements inFigure 1 and one group focus on a semiconductor supply chain reference model. The secondset of breakouts had three groups consider the interaction between various elements inFigure 1, one group talked about the incorporation of humans in the supply chain, and onediscussed how to go from the reference model to a specific semiconductor supply chain modelinstance.

The final Friday set of breakouts included three groups that discussed process models ofmultiple elements from Figure 1 and the flow of information needed between the elements toprovide core elements of a reference model. Another group discussed the role of agents ina semiconductor company’s supply chain. The final breakout group discussed the level ofdetail needed in a top down reference model. Friday consisted of a discussion on the requiredcore elements of a reference model for semiconductor supply chains and a wrap-up session.

Key Take AwaysThere were a number of key findings and areas for future research that were identified in theseminar. We will first summarize some of the key findings and will follow this with someareas for future research.

One of the first findings was that the participants generally agreed that the differentelements in Figure 1 are reasonably well understood by both the industrial and academiccommunities, but the interactions between the elements are less well understood. Havingsaid this, a number of the software solutions for the elements are not geared toward thecomplexities of the semiconductor industry (e.g. ATP/APS systems are generally focused onprofit maximization and ignore many of the system complexities). Second, it appears thatthere are still limitations in solution approaches in practice such as: capacity generally isexpressed without regard to mix; fixed lead times are generally still assumed despite researchdone on clearing functions for planning; and ignoring all but production lots when developing


Table 1 Individual Presentations

Topic Presenter

Network Planning Scott MasonDemand Planning Chen-Fu ChienCapacity Planning Adar KalirMaster Planning Thomas Ponsignon

Order Release Planning Hubert MissbauerATP José Framinán

Global and Local Decisions Stéphane Dauzère-PérèsComplexity in SCM Can Sun

Inventory Management Jei-Zheng WuSemiconductor Supply Chain Contracts Cathal Heavey

Supply Chain Planning Coordination via Planning Ton de KokSustainability in SCM Jesus Jimenez

Simulation Modeling for Supply Chains (Distributed) Leon McGinnisSimulation for Semiconductor Manufacturing (Supply Chain) Peter Lendermann

Decision MakingAgent-based Simulation Iris Lorscheid

Table 2 Breakout Sessions

Session Topic Participants (lead in bold)

1

Demand and Inventory Planning Uzsoy, de Kok, Chien, Missbauer, LeeCapacity Planning Kalir, Knopp, Lörscheid, Mönch,

Dauzère-PérèsMaster Planning Fordyce, Herding, Mason, Ovacik,

PonsignonAvailable To Promise (ATP) Framinán, Heavey, Ehm, Tirkel, Jimenez

Reference Model Rose, Sun, Weigert, Lendermann,McGinnis

2

Demand and Inventory Planning– de Kok, Tirkel, Uzsoy, Dauzère-PérèsCapacity Planning– Master Planning

Master Planning– Available to Framinan, Fordyce, Herding, Ponsignon,Promise Kalir

Master Planning– Factory Missbauer, Weigert, Jimenez, Mönch,Knopp

Incorporation of Human Behavior in Ehm, Heavey, Mason, Rose, Lörscheidthe Supply Chain

From Reference Model to Systems Chien, Sun, Lendermann, McGinnis,Ovacik

3

Process Models Demand and Kok, Tirkel, Uzsoy, Dauzère-PérèsInventory Planning– CapacityPlanning– Master PlanningProcess Models Demand and Fordyce, Herding, Kalir, Ponsignon

Inventory Planning– ATP – MasterPlanning

Process Models Master Planning– Missbauer, Weigert, Jimenez, Mönch,Factory p

Agents in the Level 3 Supply Chain Lörscheid, Mason, Ovacik, SunTop Down Reference Model Ehm, Heavey, Lendermann, McGinnis,

Rose

16062


plans. Third, as indicated above both the industrial and academic participants generallyagree that the integration of the decisions made by the different elements is often fairly adhoc and could/should be improved. Finally, the participants generally agreed that there doesnot currently exist an adequate reference model for the semiconductor supply chain. In fact,there is not even a reasonable set of data sets that describe instances of the semiconductorsupply chain such as the MIMAC datasets at the factory level. There is some indication thata reference model and incorporating human behavior of the various decision makers on thesupply chain level will help to better understand supply chains producing and containingsemiconductors.

In addition to the findings mentioned above, several areas for future research wereidentified. An overarching idea was that the future research should focus more on formulationof appropriate models because this is fundamentally more important than the actual solutiontechniques chosen. Some of the future research areas are included below:

Using event-driven process chains (EPCs) to model/visualize planning processes.Developing better integration of various decisions made in the elements of Figure 1.Combining rolling horizon strategies with demand forecast evolution models.Incorporating sustainability aspects into supply chain models.Developing stochastic model versions of current deterministic models.Incorporating the behavior of human decision makers (this will be useful, but challenging).Exploring the use of different simulation paradigms (systems dynamics, agent-based,hybrid models, reduced simulation models) to model and analyze semiconductor supplychains.

Next StepsAs a way to further the discussion of and collaboration on the topics of the seminar, Prof. LarsMönch, Prof. Chen-Fu Chien, Prof. Stéphane Dauzère-Pérès, Hans Ehm, and Prof. JohnFowler are guest editing a special issue of the International Journal of Production Research(IJPR) entitled Modeling and Analysis of Semiconductor Supply Chains. The deadline forsubmission is September 1, 2016. This date was selected to allow time for ideas created bythe participants of the seminar to be incorporated into papers for the special issue. The Callfor Papers can be found at the following address:

http://explore.tandfonline.com/cfp/est/semiconductor-supply-chains-call

Acknowledgements. The seminar organizers would like to thank Infineon Technologies AGfor their support of the seminar. The seminar also would not have been nearly as productivewithout the active contribution of every attendee, and for that the organizers are extremelygrateful.

References1 Chien, C.-F., Dauzère-Pérès, S., Ehm, H., Fowler, J. W., Jiang, Z., Krishnaswamy, S.,

Mönch, L., Uzsoy, R. (2011): Modeling and Analysis of Semiconductor Manufacturing ina Shrinking World: Challenges and Successes. European Journal of Industrial Engineering,5(3), 254–271, 2011. http://dx.doi.org/10.1504/EJIE.2011.041616

2 Mönch, L., Fowler, J. W., Mason, S. J. (2013): Production Planning and Control forSemiconductor Wafer Fabrication Facilities: Modeling, Analysis, and Systems. SpringerOperations Research/Computer Science Interfaces, New York, Vol. 52, 2013. ISBN 978-1-4899-9901-6.

http://explore.tandfonline.com/cfp/est/semiconductor-supply-chains-call

http://dx.doi.org/10.1504/EJIE.2011.041616


2 Table of Contents

SummaryChen-Fu Chien, Hans Ehm, and Lars Mönch . . . . . . . . . . . . . . . . . . . . . 28

Overview of TalksIndustry Overview – Modeling and Analysis of Semiconductor Supply ChainsHans Ehm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

Supply Chain Management Planning for the Production of Semiconductor BasedPackaged Goods Tasks, Purpose, Challenges, and a bit of History A Perspectivefrom Agents of Change in the TrenchesHarpal Singh, John Milne, Ken Fordyce, and Robert Tenga . . . . . . . . . . . . . . 38

Value Chain & Ecosystem Perspective for Modeling and Analysis of SemiconductorManufacturing & SCMChen-Fu Chien . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Master Production Scheduling Journey at IntelIrfan Ovacik . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

Modeling and Analysis of Semiconductor Supply Chains: Preliminary Results of aLiterature SurveyLars Mönch, Reha Uzsoy, and John Fowler . . . . . . . . . . . . . . . . . . . . . . 40

Network PlanningScott J. Mason . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

Modeling and Analysis of Semiconductor Manufacturing & SCM: Demand Fore-cast/Planning for Capacity PlanningChen-Fu Chien . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

Capacity Planning in Semiconductor Manufacturing: A Practical PerspectiveAdar Kalir . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

Master Planning in Semiconductor Supply ChainsThomas Ponsignon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

Optimization-based Order Release PlanningHubert Missbauer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

Available To Promise (ATP)José M. Framinán . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

Consistency between Global and Local Scheduling Decisions in SemiconductorManufacturingStéphane Dauzère-Pérès . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

Complexity Management in Semiconductor Supply ChainCan Sun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

Inventory Management in Semiconductor Supply ChainsJei-Zheng Wu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

Semiconductor Supply Chain – ContractsCathal Heavey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

16062


Coordinating the Semiconductor Supply Chain by PlanningTon de Kok . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

Sustainability in Supply Chain ManagementJesus Jimenez . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

What Is a Reference Model and What Is It Good For?Leon F. McGinnis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

(Distributed) Simulation for Semiconductor Manufacturing (Supply Chain) Decision-MakingPeter Lendermann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

Agent-based SimulationIris Lorscheid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

Breakout ReportsDemand and Inventory PlanningReha Uzsoy, Ton de Kok, Chen-Fu Chien, Hubert Missbauer, and Peng-Chie Lee . 47

Capacity PlanningAdar Kalir, Sebastian Knopp, Iris Lörscheid, Lars Mönch, and Stéphane Dauzère-Pérès 48

Master PlanningKenneth Fordyce, Raphael Herding, Scott Mason, Irfan Ovacik, and Thomas Ponsignon 50

ATPJosé Framinán, Cathal Heavey, Hans Ehm, Israel Tirkel, and Jesus Jimenez . . . . 51

Reference ModelOliver Rose, Can Sun, Gerald Weigert, Peter Lendermann, and Leon McGinnis . . 52

Demand & Inventory Planning – Capacity Planning – Master PlanningTon de Kok, Israel Tirkel, Reha Uzsoy, and Stéphane Dauzère-Pérès . . . . . . . . 53

MP & ATPJosé Framinán, Ken Fordyce, Raphael Herding, Thomas Ponsignon, and Adar Kalir 54

MP & FactoryHubert Missbauer, Gerald Weigert, Jesus Jimenez, Lars Mönch, and Sebastian Knopp 56

Incorporation of Human Behavior in the Supply ChainHans Ehm, Cathal Heavey, Scott J. Mason, Oliver Rose, Iris Lorscheid . . . . . . 57

From Reference Model to Systems (the SC System in 20201)Chen-Fu Chien, Can Sun, Peter Lendermann, Leon F. McGinnis, and Irfan Ovacik 58

Process Models Demand and Inventory Planning, Capacity Planning, MasterPlanningTon de Kok, Israel Tirkel, Reha Uzsoy, and Stéphane Dauzère-Pérès . . . . . . . . 59

Process Models Demand and Inventory Planning, ATP, Master PlanningKenneth Fordyce, Raphael Herding, Adar Kalir, and Thomas Ponsignon . . . . . . 60

Process Models Master Planning and FactoryHubert Missbauer, Gerald Weigert, Jesus Jimenez, Lars Mönch, and Sebastian Knopp 61

Agents in the Level 3 Supply ChainIris Lorscheid, Scott Mason, Irfan Ovacik, and Can Sun . . . . . . . . . . . . . . . 61


Top Down Reference ModelHans Ehm, Cathal Heavey, Peter Lendermann, Leon McGinnis, and Oliver Rose . 62

Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

16062


3 Overview of Talks

3.1 Industry Overview – Modeling and Analysis of SemiconductorSupply Chains

Hans Ehm (Infineon Technologies – München, DE)

License Creative Commons BY 3.0 Unported license© Hans Ehm

The semiconductor innovation race continues, comparing with 6 years ago, many new tech-nologies emerge, which also bring us a lot of challenges. The semiconductor supply chain ischaracterized by steep ramps, short product life cycles and long cycle times. Especially forcompanies who are far away from the end customers, the bullwhip effect is more significantand has a huge impact on the supply chain complexity. This speech discusses from Infineon,a company point of view and shares its best practices as well as its challenges on productionplanning and supply chain management. On one hand, Globalization and Flexibility areeffective tools to manage global supply chain; on the other hand, the complexity of manu-facturing process and planning is thus increased. The Modeling and simulation techniquesbecome an important lever to solve complex problems and support decision making. Afour-level approach is proposed: from the tool and work center bottom level to the end-to-endsupply chain top level. A semiconductor supply chain simulation library called SCSC-SIMLIBwas thus designed together with academia and implemented, and key objects in each levelare developed. Following this structure many concrete examples and various applicationswithin Infineon are given and demonstrated. Finally, the hot topics, e.g., human behaviorsanalysis, disruption management and change management are highlighted; and the futuretrends towards the Industry 4.0, IoT, and big data are referred to.

3.2 Supply Chain Management Planning for the Production ofSemiconductor Based Packaged Goods Tasks, Purpose,Challenges, and a bit of History A Perspective from Agents ofChange in the Trenches

Harpal Singh, John Milne, Ken Fordyce, and Robert Tenga (Arkieva – Wilmington, US)

License Creative Commons BY 3.0 Unported license© Harpal Singh, John Milne, Ken Fordyce, and Robert Tenga

Joint work of Peter Lyon, Gary Sullivan, Alfred Degobotse, Brian Denton, Bob Orzell

The purpose of any demand supply network is to meet prioritized demand on time withoutviolating constraints and as much as possible meet business policies (inventory, preferredsuppliers, request and commit date, etc.). Typically, the demand supply network for theproduction of semiconductor based packaged goods (SBPG) is divided into FAB and POST-FAB. The dynamic interaction between the two is limited in nature for logical and historicalreasons. One reason is the nature of complexity which makes life interesting for planners isdifferent between FAB and POST-FAB. FAB has long routes, reentrant flow, deploymentand the ever present shadow of the operating curve – to name a few – generating wafer start/ cycle time focus. POST-FAB is faced with constant exit demand uncertainty, allocation ofshared components and capacity to competing demands, alternative operations; transportdecisions, and the all-important “plan repair” – name a few – generating an exit demand /






efficiency frontier focus. Their differences become clear when examining the nature of themodels (from spreadsheets to optimization) supporting decisions and analysis. The purposeof this presentation was to provide overview of the current best practices for central planningand identify computational challenges to reduce the current slack and make these networksmore responsive.

3.3 Value Chain & Ecosystem Perspective for Modeling and Analysisof Semiconductor Manufacturing & SCM

Chen-Fu Chien (National Tsing Hua University, TW)

License Creative Commons BY 3.0 Unported license© Chen-Fu Chien

On the basis of my extensive collaborative studies with high-tech industries and the trendsI have observed, this talk aims to address emerging research issues, from value chain &ecosystem perspective, driven by the needs in modeling and decision analysis in manufacturingthat is enabled by the advances of automation and information technology. Semiconductormanufacturing is one of the most complicated and capital-intensive industries that is drivenby Moore’s Law for continuous improvements for technology advance and cost reduction.Therefore, high-tech companies are confronting with various decision problems involved instrategy, manufacturing, and technology that are characterized by uncertain (incomplete ormassive) information and a need for tradeoff among different objectives and justification forthe decisions to align with the overall strategic objectives. In the fully automation productionfacility such as semiconductor fab, manufacturing intelligence approaches have been developedto model the right decision problems involved in the operations and manufacturing strategiesand to extract useful information to estimate the parameters and derive decision rules via datamining to enhance decision quality and production effectiveness. Owing to the advancementof information technology, researchers have developed various data mining methodologies toextract potentially useful patterns through semi-automatic exploration of a huge databasein various domains. This talk will also use a number of empirical studies to illustrate theobservation and the needs. Finally, we conclude this talk with discussion of the ongoingchanges of manufacturing in high-tech industries and the emerging research directions. Inaddition, since the uncertainty involved in demand forecast is increasingly amplified withthe forecast lead-time, high-tech companies often suffer the risks of oversupply and shortageof capacity that will affect the profitability and growth. High-tech industries includingsemiconductor and TFT-LCD industries are capital intensive, in which the capacity planand corresponding capital investment decisions are critical due to demand fluctuation. Oncethe capacity is planned, the company may suffer the risks of either low capital-effectivenessdue to low capacity utilization and capacity oversupply, or poor customer satisfaction causedby the capacity shortage. Most of the existing studies focused on solving the long-termcapacity shortage issue through optimizing the capacity investment plan, or medium-termcapacity plan to allocate demands among the wafer fabrication facilities (fabs) to balancethe loading and product mix. Focusing on a real setting, this talk aims to share a proposedsystematic decision method to analyze short-term solutions of cross-company capacity backupbetween the companies in the semiconductor industry ecosystem. In particular, a gametheory and decision tree analysis model was developed to support this decision. A case studywas conducted with real data of semiconductor manufacturing companies in Taiwan for

16062




validation. The results have demonstrated practical viability of this approach. The approachsuggested has been implemented in this company. This talks concludes with a case studyon the paradigm shifts of Global Unichip to address the issues involved in value chain &ecosystem perspective.

3.4 Master Production Scheduling Journey at IntelIrfan Ovacik (Intel Corporation – Chandler, US)

License Creative Commons BY 3.0 Unported license© Irfan Ovacik

In this talk, we discuss the journey that Intel took to build a world-class Master ProductionScheduling (MPS) solution. MPS involves decisions as to what each factory in the supplynetwork needs to manufacture, when, and at what quantities so as to meet the demandwhile minimizing supply chain costs. Most semiconductor companies have elected to usea best-in-class strategy and partnered with solution providers to build their supply chainplanning solutions. In contrast, Intel chose to develop its solution in-house. We discuss howIntel decomposed the problem into smaller problems to better align with its strategy andorganization boundaries and share the lessons learned during this journey.

3.5 Modeling and Analysis of Semiconductor Supply Chains:Preliminary Results of a Literature Survey

Lars Mönch (FernUniversität in Hagen, DE), Reha Uzsoy (North Carolina State University,US), and John Fowler (Arizona State University, US)

License Creative Commons BY 3.0 Unported license© Lars Mönch, Reha Uzsoy, and John Fowler

We conduct and report on a literature review of academic and industrial research in thedomain of supply chain management in the semiconductor sector. Areas examined includedemand planning, inventory management, network design, master planning, productionplanning, contracts and coordination, supply chain simulation, and involved Informationsystems. We also briefly discussed research from other domains and future research directions.

3.6 Network PlanningScott J. Mason (Clemson University, US)

License Creative Commons BY 3.0 Unported license© Scott J. Mason

Semiconductor network planning is a challenging problem of great importance. However, notall plans are created equal – a variety of factors that can influence a plan include perspective,level of granularity, objective function(s), and whether or not uncertainty is included. Wediscuss an industrial case study using both Excel and mathematical optimization to analyzenetwork plans for wafer starts, assembly starts, and test starts to minimize total costs.








We examine target inventory levels, lead times, and capacity constraints to illustrate keytrade-offs faced by semiconductor manufacturers.

3.7 Modeling and Analysis of Semiconductor Manufacturing & SCM:Demand Forecast/Planning for Capacity Planning

Chen-Fu Chien (National Tsing Hua University, TW)

License Creative Commons BY 3.0 Unported license© Chen-Fu Chien

Semiconductor industry is capital intensive in which capacity utilization significantly affectthe capital effectiveness and profitability of semiconductor manufacturing companies. Thus,demand forecasting provides critical input to support the decisions of capacity planningand the associated capital investments for capacity expansion that require long lead-time.However, the involved uncertainty in demand and the fluctuation of semiconductor supplychains make the present problem increasingly difficult due to diversifying product lines andshortening product life cycle in the consumer electronics era. Semiconductor companies mustforecast future demand to provide the basis for supply chain strategic decisions includingnew fab construction, technology migration, capacity transformation and expansion, toolprocurement, and outsourcing. Focused on realistic needs for manufacturing intelligence,this talk aims to share a proposed multi-generation diffusion model for semiconductorproduct demand forecast, namely the SMPRT model, incorporating seasonal factor (S),market growth rate (M), price (P), repeat purchases (R), technology substitution (T),in which the nonlinear least square method is employed for parameter estimation. Anempirical study was conducted in a leading semiconductor foundry in Hsinchu SciencePark and the results validated the practical viability of the proposed model. Furthermore,the forecasted demands can be used for capacity planning. Due to constant technologyadvance driven by Moore’s Law in semiconductor industry, multiple production technologiesgenerally co-exist in a wafer fabrication facility with utilization of a pool of common tools formultiple technologies and critical tools dedicated for a specific technology. In semiconductorindustry, demand forecasts are rolling and updated when the latest market and demandinformation is available. This demand forecast mechanism makes forecast errors in differenttime periods correlated. Because part of the equipment is common for products of differenttechnologies, production managers have limited flexibility to dynamically allocate the capacityamong the technologies via capacity migration. The possibility of capacity migration andinterrelationship among different technologies make capacity planning difficult under demandand product-mix uncertainties. We developed a dynamic optimization method that capturesthe unique characteristics of rolling demand forecast mechanism to solve capacity expansionand migration planning problems in semiconductor industry. We also proposed a mini-maxregret strategy for capacity planning under risk. We estimate the validity and robustnessof the proposed dynamic optimization method in an empirical study in a semiconductormanufacturing company in Taiwan. The results showed practical viability of this approachand the findings can provide useful guidelines for capacity planning process under rollingforecast mechanism.

16062




3.8 Capacity Planning in Semiconductor Manufacturing: A PracticalPerspective

Adar Kalir (Intel Israel – Qiriat-Gat, IL)

License Creative Commons BY 3.0 Unported license© Adar Kalir

Capacity planning problems have been researched extensively over the past two decadesand still are, even with more intensity, in recent years. In this talk, an industry perspectiveis provided on the classification of the problems and the evolution in the various solutionapproaches to these problems. Insight to current and future challenges is also discussed withrespect to both strategic and tactical capacity planning.

3.9 Master Planning in Semiconductor Supply ChainsThomas Ponsignon (Infineon Technologies – München, DE)

License Creative Commons BY 3.0 Unported license© Thomas Ponsignon

Given the specifics of semiconductor manufacturing networks, the development of enterprise-wide planning approaches that are computationally tractable and address the uncertaintiestypically encountered in this industry remains particularly challenging. This talk focuses onMaster Planning in semiconductor supply chains. Master Planning determines productionquantities of end-products in a manufacturing network to meet external demands (e.g.,customer orders and forecasts) and internal demands (e.g., requests for stock replenishment).A mid-term horizon (i.e., six months) expressed in weeks is considered. The outcome iscapacitated production requests. Master Plan details the aggregated Sales & OperationsPlan and it is the main input for Site Scheduling and Order Promising. In this presentation,Master Planning formulations and solving approaches as found in the scientific literature andthe industry are discussed. Their performances with regard to solution quality, computationalburden, and plan stability are outlined. Some insights from a semiconductor manufacturerinto a real-world Master Planning approach are showed. Finally, trends and future challengesboth for academia and practitioners are presented.

3.10 Optimization-based Order Release PlanningHubert Missbauer (Universität Innsbruck, AT)

License Creative Commons BY 3.0 Unported license© Hubert Missbauer

Order release is the interface between the centralized planning of the material flow throughthe entire logistic chain and detailed scheduling of the work orders within the production units.The presentation deals with multi-period models that optimize order release quantities perproduct and period based on a descriptive model of the material flow within the productionunit that is represented as a network of work centers. We describe models with fixed targetlead times as well as models with load-dependent lead times. Their shortcomings and therelevant research issues are outlined.








3.11 Available To Promise (ATP)José M. Framinán (University of Sevilla, ES)

License Creative Commons BY 3.0 Unported license© José M. Framinán

Available-To-Promise (ATP) systems deal with a number of managerial decisions relatedto Order Capture activities in a company, including order acceptance/rejection, due datesetting, and resource scheduling. These different but interrelated decisions have been oftenstudied in an isolated manner, and even the terminology and models employed differ largely.This communication intends to give an overview of the main contributions in the field andpresent some open issues for discussion.

3.12 Consistency between Global and Local Scheduling Decisions inSemiconductor Manufacturing

Stéphane Dauzère-Pérès (École des Mines de Saint-Etienne, FR)

License Creative Commons BY 3.0 Unported license© Stéphane Dauzère-Pérès

Joint work of R. Sadeghi and S. Dauzère-Pérès

The operational level in semiconductor manufacturing can be divided into global (fab) andlocal (workshop) decision levels. The global level provides objectives or constraints for thelocal level. Based on previous research, our goal is to develop a framework to simultaneouslyoptimize multiple performance measures and support the consistency between global andlocal scheduling decisions. A first approach is based on a global rule that aims at ensuringthat lots satisfy their time constraints. Numerical experiments on industrial data show theinterest of the approach. A first optimization model has also been proposed to take intoaccount multiple constraints and objectives.

3.13 Complexity Management in Semiconductor Supply ChainCan Sun (Infineon Technologies – München, DE)

License Creative Commons BY 3.0 Unported license© Can Sun

Semiconductor supply chain is a complex system characterized by increasing number ofinteractive and interdependent components with dynamic behaviors working together as anentirety. Many innovative activities occur in the daily supply chain and manufacturing, andthese changes inevitably bring in the complexities to the organization. But not all of themare valuable to the business goals. Decision makers want to keep value-added complexity andreduce non-value-added complexity. However, the quantitative analysis of the complexitygenerated by these components and their behaviors and its impact on the system still lackspractical methodology. Therefore we design a framework to measure the complexity ofSemiconductor supply chain. The first step is to understand and represent the complexityusing an conceptual model called PROS (process, role, object, state) idea, which providesan understandable and structural way to describe the complexity.., we can thus develop theformulas to measure system complexity based on the metrics of process complexity as well

16062








as the properties of complex system. A simplified small real example from semiconductorsupply chain is used to demonstrate this approach. It is also noticed that the human playsan important role in the complexity due to its uncertainty behaviors. This is demonstratedand investigated through a classical supply chain phenomenon – the bullwhip effect, whichcan be demonstrated using a serious game called beer distribution game.

3.14 Inventory Management in Semiconductor Supply ChainsJei-Zheng Wu (Soochow University, TW)

License Creative Commons BY 3.0 Unported license© Jei-Zheng Wu

Joint work of C.-F. Chien, J.-Z. Wu, and H.-C. Yu

The rapid technology development and shortening product life cycle lead to high risk ofproduct obsolescence. Manufacturers still need to hold a reasonable level of inventory tosatisfy customers under demand uncertainty and long lead-time. This study introducesthe inventory days into the multi-stage inventory model that incorporates with complexBOM, product substitution, wafer release schedules (wafer start), turnaround times (TAT,lead times), production plans, safety stock strategies, and end product-demand forecastswith four banks, i.e., VIA Bank, Wafer Bank, Die Bank, Finished-Good Bank for delayeddifferentiation and postponement strategy. Survival analysis is used to estimate inventorydays and to group products. Inventory ages and accrued provision rates are also added intothe model for the compliance with international financial reporting standards, No. 2.

References1 Wu, J.-Z., Yu, H.-C., and Chien, C.-F. (2014/12), “Inventory survival analysis for semicon-

ductor memory manufacturing,” Proceedings of Winter Simulation Conference: Modelingand Analysis of Semiconductor Manufacturing (MASM), Savannah, pp. 2591–2599, GA,USA, December 7-10.

2 Wu, J.-Z., Chien, C.-F. and Tsou, Y.-C. (2014/08), “Multistage semiconductor memoryinventory model based on survival analysis,” IEEE International Conference on AutomationScience and Engineering, pp. 613–618, Taipei, Taiwan, August 18-22.

3 Wu, J.-Z. (2013/05), “Inventory write-down prediction for semiconductor manufacturingconsidering inventory age, accounting principle, and product structure with real settings,”Computers & Industrial Engineering, 65(1), pp. 128–136.

3.15 Semiconductor Supply Chain – ContractsCathal Heavey (University of Limerick, IE)

License Creative Commons BY 3.0 Unported license© Cathal Heavey

This presentation presented an introduction to contracts used in Semiconductor SupplyChain. It first presented the purpose of contracts, which are to coordinate a supply chain,the flow of product, information and funds. The presentation then stated that contracts needto be included into supply chain planning as they are a core element of planning in a similarway that inventory control is. The presentation also presented information on optimizing aRHF contract for a semiconductor SC with forecast error. It then presented an introduction






to comparing option contracts with RHF contracts. The presentation finally stated that thisis an important aspect of semiconductor SC.

3.16 Coordinating the Semiconductor Supply Chain by PlanningTon de Kok (TU Eindhoven, NL)

License Creative Commons BY 3.0 Unported license© Ton de Kok

Starting from the Eindhoven Framework for Production and Inventory Control (EFPIC) wediscuss decision hierarchies and mathematical models used at different levels. The key featureof the hierarchy proposed is the (planned) lead time that enables decomposition betweensupply chain planning and production unit planning and scheduling. Though stochasticmulti-item multi-echelon inventory models produce empirically valid results, they cannot copewith the planning complexities of the semiconductor supply chain. This leads to alternativemathematical programming formulations. We briefly discuss the interplay between APS andhuman planner and scheduler.

3.17 Sustainability in Supply Chain ManagementJesus Jimenez (Texas State University – San Marcos, US)

License Creative Commons BY 3.0 Unported license© Jesus Jimenez

Joint work of Tongdan JinMain reference V. Santana-Viera, J.A. Jimenez, T. Jin, and J. Espiritu, “Implementing Factory Demand Response

with On-site Renewable Energy: A Design-of-experiment Approach”, in International Journal ofProduction Research – Special Issue on Energy-aware Manufacturing Operations, Vol. 53(23),pp. 7034–7048, 2014.

URL http://dx.doi.org/10.1080/00207543.2014.957877Main reference S. Villarreal, J.A. Jimenez, T. Jin, and M. Cabrera-Rios, “Designing a Sustainable and Distributed

Generation System for Semiconductor Wafer Fabs”, in IEEE Transactions on Automation Scienceand Engineering, Vol. 10 (1), pp. 16–26, 2012.

URL http://dx.doi.org/10.1109/TASE.2012.2214438Main reference L. Sanders, S. Lopez, G. Guzman, J.A. Jimenez, and T. Jin, “Simulation of a green wafer fab

featuring solar photovoltaic technology and storage system”, in Proc. of the 2012 WinterSimulation Conference (WSC), pp. 1–12, 2012.

URL http://dx.doi.org/10.1109/WSC.2012.6465269

The high-tech facilities used for the fabrication of semiconductor wafers consume a significantamount of electricity. The impact of energy consumption on climate change and the risingcost of energy have become a challenging issue facing the semiconductor manufacturingindustry today. ITRS urges chip manufacturers to reduce carbon footprints by designingand deploying green and sustainable manufacturing facilities. The focus of this presentationis to present opportunities for the supply chain in semiconductor manufacturing. We studiedthe penetration of renewable technology in wafer fabs and identified the costs of thesesystems. We measure carbon emissions and probability of black outs. Future research is thedevelopment of models that measure carbon dioxide savings across the supply chain.

16062





http://dx.doi.org/10.1080/00207543.2014.957877

http://dx.doi.org/10.1080/00207543.2014.957877

http://dx.doi.org/10.1080/00207543.2014.957877

http://dx.doi.org/10.1080/00207543.2014.957877

http://dx.doi.org/10.1080/00207543.2014.957877

http://dx.doi.org/10.1109/TASE.2012.2214438




http://dx.doi.org/10.1109/WSC.2012.6465269





3.18 What Is a Reference Model and What Is It Good For?Leon F. McGinnis (Georgia Institute of Technology, US)

License Creative Commons BY 3.0 Unported license© Leon F. McGinnis

Semiconductor supply chains are very complex, distributed, and dynamic systems, and it issimply not possible to make important decisions about designing, planning, or controllingthem without computational decision support. Today, a grand challenge is to make thatcomputational support as ubiquitous as using Google Maps on your mobile phone to findthe short route from Frankfurt to Schloss Dagstuhl. In the status quo, however, thesecomputational decision support tools require a great deal of customization for every differentdecision maker. So how do we change that?

An approach that holds great promise involves adapting two key concepts from computerscience. The first is meta-modeling, or the creation of reference models of the domain ofinterest. These reference models can be articulated using languages like UML, its variant,SysML, or related Ecore, and define the semantics and to some extent the syntax for describinginstances of problems in the semiconductor supply chain. The second is model-to-modeltransformation, which allows us to capture the “algorithm” for translating one model – say aformal model of a problem instance in the semiconductor supply chain domain – into anothermodel – say a large scale optimization for planning the production and logistics for devicemanufacture over the next six months. Once these models are generated, there are manysolvers that can be used effectively to compute solutions.

This presentation argues for the importance of meta-modeling and model-to-modeltransformation as keys for bringing academic research results into semiconductor supplychain practice faster, and more reliably.

3.19 (Distributed) Simulation for Semiconductor Manufacturing(Supply Chain) Decision-Making

Peter Lendermann (D-SIMLAB – Singapore, SG)

License Creative Commons BY 3.0 Unported license© Peter Lendermann

The presentation describes a number of challenges associated with the use of distributedsimulation for decision-making in semiconductor supply chain management such as the hetero-geneity of external drivers of such supply chains, the need of representing planning decisionsand human decision-making as well as the constraints arising from time synchronizationbetween the different federates of such a simulation system.

Additional challenges are arising from the fact that in a real fab environment product mixchanges continuously and operations are never in steady state. To address this the author isadvocating a simulation-based WIP Management approach, allowing a great reduction ofvariability in fab operations. One of the current research questions is whether this kind ofvariability reduction can also be reached on the Semiconductor Supply Chain level. Thiswould obviously depend on which specific use cases are relevant on the Supply Chain level andwhich regular SCM decisions are to be enabled. As of now the within-echelon SemiconductorSCM challenge, i.e. the Borderless Fab, appears to be the most promising next application.






A number of key learnings have evolved from the author’s experience with real-worldsemiconductor manufacturing systems and supply chains: In particular, there is no need tofind an optimum because a real optimum can only be found with a perfect model. In realityit is sufficient to find a “considerably better” solution as fast (i.e. with as few iterations)as possible. This can be achieved through “smart” heuristics and enabled by a parallelcomputing infrastructure. It is essential though that constraints are portrayed as much aspossible because otherwise “solutions” will be generated that are infeasible in practice andmanagers will lose trust in the enabling software tool. For this reason the Discrete EventSimulation approach can be quite powerful, however, automation is key not only for “analysis”but also for “modelling” including model verification and model maintenance as well as datacalibration. Otherwise it would not be possible neither to keep up with the fast-changing realworld operations nor to carry out “routine” analysis instantly and at zero (variable) cost.

3.20 Agent-based SimulationIris Lorscheid (TU Hamburg-Harburg, DE)

License Creative Commons BY 3.0 Unported license© Iris Lorscheid

Agent-based simulation may provide new perspectives on supply chains from three angles:analyzing the supply chain as an emergent system resulting from individual interactions,incorporating the uncertainties caused by human behavior, and using artificial agents tofind new designs or strategies. The first perspective models individual (inter-)actions incomplex systems and analyzes the resulting system behavior. This provides an understandingon self-organizing processes and emergent phenomena that are not explicitly modeled oreven understood by the modeler. By analyzing the human factor in supply chain planningprocesses, the second perspective, good or bad individual strategies may be observed, theireffect analyzed, and the decision maker environment designed in a way that those arepromoted or prevented. Artificial agents, finally, may help to optimize complex system byoptimizing individual strategies by means of machine learning algorithms.

4 Breakout Reports

4.1 Demand and Inventory PlanningReha Uzsoy, Ton de Kok, Chen-Fu Chien, Hubert Missbauer, and Peng-Chie Lee

License Creative Commons BY 3.0 Unported license© Reha Uzsoy, Ton de Kok, Chen-Fu Chien, Hubert Missbauer, and Peng-Chie Lee

The purpose of demand and inventory planning is to create reliable forecasts of demand forthe portfolio of products offered by a semiconductor manufacturer and to determine theamount of inventory needed to buffer against the inevitable demand uncertainty. Effectivedemand planning can guide companies to improve the accuracy of revenue forecasts, aligninventory levels with peaks and troughs in demand, and enhance profitability. In today’senvironment, utilization is the “name of the game” since fab investment is the main costcomponent, but is mostly a sunk cost and has a long lead time to add capacity. Backendutilization is also important, but is generally not quite as important because the equipment

16062






is not as expensive and has shorter lead times to purchase. In addition, many companiessubcontract backend operations. Overall, there are low marginal costs to meet demand andtherefore profit margins are often relatively high for satisfying additional demand.

As seen in Figure 1, the output of demand planning provides inputs to both networkplanning (strategic needs) and capacity planning (tactical needs) in regards to likely futuredemand. It also provides inputs to the available to promise (ATP) system in terms demandexpected in the near term. In some companies (such as TSMC), the demand planner usesdata and models along with tacit knowledge to act as a middle man in consolidating regionalsales plans, correcting for correlations between products, and in communicating with capacityplanning.

There are a number of things that need to be done to improve the overall effectiveness ofdemand and inventory planning. First, it would be helpful to develop swim lane schemesof the relation between sales planning, demand planning and capacity planning. Second,even though utilization determines long-term planning, the short-term match of supply anddemand requires new ATP functionality, in particular the need for better allocation schemesduring tight demand is very important. Third, there is a need to better integrate productionplanning and inventory planning. As researchers investigate these issues, the focus shouldbe on problem formulation, not on algorithm development and should keep in mind thatdetermining the correct cost structure is key to arrive at the right solution. Finally, there isa need to distinguish between fabless companies, foundries and fab owners.

4.2 Capacity PlanningAdar Kalir, Sebastian Knopp, Iris Lörscheid, Lars Mönch, and Stéphane Dauzère-Pérès

License Creative Commons BY 3.0 Unported license© Adar Kalir, Sebastian Knopp, Iris Lörscheid, Lars Mönch, and Stéphane Dauzère-Pérès

Capacity planning incorporates planning for future capacity to ensure that capacity isavailable for production planning (tactical) and adjusted for forecasted demand (strategic).Capacity planning is fed from the aggregate planning and feeds into the master planning.The following decisions are addressed by capacity planning:

equipment changes (buy, convert, qualify)feasibility check for the production plan (capacity-wise, not schedule)wafer start quantities per period by product (aggregate, not fab level)performance expected (production cycle time)

Figure 2 shows where capacity planning is located in the production planning and schedulinghierarchy.

The following objective function is maximized:profit (revenue – cost)cost (wafer cost, unit cost, backlog and inventory cost)

Risk minimization is also considered. The different types of capacity planning problems aresummarized in Figure 3.

The following input data is used for capacity planning:planning horizon and time bucketdemand plan (after smoothing) per product over planning horizoncurrent equipment, and capacity by fab




Figure 2 Capacity planning in the planning and scheduling hierarchy.

Figure 3 Mapping of Capacity Planning Problems.

equipment data (RR, operations, . . . – per product)capacity factors (dedications; setups; batching; . . . )financial data (equipment value, wafer cost per fab)budget for expansion

The interface with network/demand planning is demand forecasting by product over time.The impact of multi-fab capacity is not considered. Typical modeling types are Excelfor static considerations, MILP for deterministic settings, and simulation and stochasticprogramming for a stochastic setting.

Next, limitations of the current state and future needs are discussed. The hierarchyenforces ‘independent’ problems. However, capacity planning must be considered in thenetwork/demand planning phase. Another imitation is the limited model accuracy because oftime bucketing, the estimation of operational decisions such as setup, batching, QT, storageand transportation (especially for older manual fabs), reticles, etc., the deterministic natureof the forecast (point forecasts vs. the real dynamic nature of actual demand) and sensitivityto the impact of product mix. Current approaches are limited to production lots, oftenignoring all other type of non productive wafer (NPW), engineering activities, new productintroduction (NPI). The quality of the input data (typically aggregated) is crucial, this isespecially true for scenarios for stochastic planning. APC/AEC and dedication schemes are

16062


not captured at all yet. Capacity planning is not adapted to the evolving SC in semiconductorindustry. The dynamics of changing production across fab locations, cross-processing betweenfab locations, etc is not considered. A consideration of the impact of multi-fab capacityis important. In future modeling, agent-based simulation, game theory, and nonlinearprogramming are important ingredients for addressing new complexities. Moreover, rollinghorizon approach are expected to be desirable to model decision-making changing over time.

The following issues have to be addressed in future research. From an industry point ofview, considering NPW, engineering activities, and NPI in capacity planning is important.Moreover, getting refined model accuracy is desirable. The impact of multi-fab capacity hasto be considered. From an academic point of view, getting more accurate models is one goal.Rolling horizon approach, capacity planning for the evolving SC, and considering the impactof multi-fab capacity are important future directions for academic research.

4.3 Master PlanningKenneth Fordyce, Raphael Herding, Scott Mason, Irfan Ovacik, and Thomas Ponsignon

License Creative Commons BY 3.0 Unported license© Kenneth Fordyce, Raphael Herding, Scott Mason, Irfan Ovacik, and Thomas Ponsignon

Definition and Scope

Master Planning determines production quantities of end-products in a manufacturingnetwork to meet external demands (e.g., orders and forecasts) and internal demands (e.g.,requests for stock replenishment). A mid-term horizon (i.e., six months) expressed in weeksis considered. The outcome is capacitated production requests. The master plan details thesales & operations plan and it is the main input for site scheduling and order promising.

Current state-of-the-art

Most semiconductor manufacturing companies blend commercial solutions (e.g., JDA) andhome-grown solutions (e.g., Intel). Regardless of the approach, most firms have operationsresearch experts for development and analysis. The solution approaches typically involve solv-ers, which may be based on optimization techniques and/or rules-based heuristic approaches(e.g., forward-and-backward demand-supply assignment). A combination of optimizationtechniques and heuristic methods is frequently used in the industry. Software developers arekey for the implementation and the deployment within a company (e.g., UI, data integration).

Inputs to Master Planning

Master planning requires the following inputs: the structure of the supply chain, demandsignal, information on capacity, current inventory and work-in-progress levels, safety stocktargets and policies, and the existing or previously generated master plan. The structure ofthe supply chain includes details on which products can be produced where, bill-of-materialinformation, geographical information, details on available equipment as well as planningparameters such as lead times and planned yields. The demand signal is expressed as pointestimates and capacity is often stated in gross terms (e.g., total starts per technology perweek, hours of loading on a tool group).




Outputs from Master Planning

Master planning provides expectations or targets of what is required from each factory overtime. It also states what each factory is expected to receive over time and from where.These outputs can be summarized as the current projected supply, which is provided toAvailable-to-Promise (ATP) usage and production planning processes. In some firms, themaster plan may be a starting point for manual adjustments.

Limitations

Capacity statements are often given without consideration of mix. Furthermore, no commonstandard is available for describing capacity. The assumption of fixed lead times are here tostay as they are often politically influenced. This leads to input factors being conservativelystated. Deterministic point estimates do not capture the uncertainty of the base system.Hence, safety stock buffers for yield and demand variations may be overstated. As a result,potential opportunities may be lost. The interpretability remains a major limitation especiallywhen optimization techniques are used to solve the master planning problem. Besides, littleis known about the recommended plan granularity and model accuracy (e.g., time bucketspecification).

Future steps

Directions for further research include: investigating which solution procedure works best inwhich situation, examining the usage of blended optimization and heuristic solution methods,Incorporating risk assessment into master planning models, and tightening model integrationwith data sources and other models.

4.4 ATPJosé Framinán, Cathal Heavey, Hans Ehm, Israel Tirkel, and Jesus Jimenez

License Creative Commons BY 3.0 Unported license© José Framinán, Cathal Heavey, Hans Ehm, Israel Tirkel, and Jesus Jimenez

ATP it is all about confirming customer orders. The product the customer/channel isordering determines the granularity. A “contract” with customer, the Operations, one initialnegotiated price, and the heuristic of the implemented APS are prerequisites.

Continuously updated capacity, inventory, committed forecast/order, requests for con-firmation are important input data. Here, capacity means supply in a two step process wheresupply is generated from capacity.

The following activities are required:check Inventorycheck Supplyuse Alternativescontinuous assessment of confirmation process.

Response to forecast/order and rigger for capacity increase are outputs. Next, the currentsituation in companies is discussed. ATP usage systems (most of the time called APS anddeveloped outside the semiconductor industry) are in place. Online data for the ATP usagesystem is already available at some semiconductor companies. ATP usage systems used in

16062




Figure 4 ATP Usage System.

the semiconductor industry are far from optimum. APS systems are optimized towards profitoptimization and lack capacity/supply complexity need of the semiconductor industry. Thefull usage of resources is key for semiconductor companies. The main functionality of anATP usage system is summarized in Figure 4.

The following future needs are identified:Make all data needed online available for an ATP usage system (at some companiesavailable).Optimize the resources, and optimize the schedule (the heuristics in the ATP usagesystem).Make use out of the profit optimization in available ATP usage systems, i.e. short termbenefit, mid term benefit, long term benefit.Measure the benefit.Measure the stability of the system.

4.5 Reference ModelOliver Rose, Can Sun, Gerald Weigert, Peter Lendermann, and Leon McGinnis

License Creative Commons BY 3.0 Unported license© Oliver Rose, Can Sun, Gerald Weigert, Peter Lendermann, and Leon McGinnis

The question what is a reference model is discussed first. The MIMAC data sets areinstances that conform to some (unstated) reference model. A domain specific language(DSL) conforms to a reference model. A DSL is specified by symbols and rules and has awell-defined semantics. SCOR is “part of” a reference model, however, there are possibleissues with precision of semantics definitions.

Next, the state of the art for supply chains is discussed. SCOR is more like an agreementhow to present KPIs, it has a very simple process model. SAP (ARIS), Oracle are used forcompany-specific SC software design, but is not public. There exist various custom models.




Very general approaches are offered by OMG that are mainly from/for computer science.There are tools such as MOF, UML, SysML, BPML.

The following limitations exist. The motivation for developing a reference model is withrespect to ROI not clear. Appropriate tools and tool chains do not exist or have somelimitations. Some inertia with respect to reference modeling can be observed in organizations.The awareness of reference modeling issues by possible stakeholders is limited.

A SCSC DSL and tools (mock-ups) are identified as possible output of future researchactivities. This includes appropriate demos. The focus is on the internal supply chain (thirdlevel). Applying for Horizon 2020 projects (maybe only as a hidden agenda of a project)might be a future step. A core working group/task force (driven by industry needs) is highlydesirable. A conference track or a journal special issue can be organized.

4.6 Demand & Inventory Planning – Capacity Planning – MasterPlanning

Ton de Kok, Israel Tirkel, Reha Uzsoy, and Stéphane Dauzère-Pérès

License Creative Commons BY 3.0 Unported license© Ton de Kok, Israel Tirkel, Reha Uzsoy, and Stéphane Dauzère-Pérès

Figure 5 shows the connections between the three planning modules, where capacity planningis actually divided into medium-term and long-term capacity planning.

Table 3 specifies various elements that characterize the planning modules.Some comments are given below that are important in understanding how these planning

decisions are performed:The role definition (in particular incentives), behavior and expertise of planners stronglyimpact the decision process,Delays (to get information and take decisions) are often non negligible (hours or days)and prevent the decision process to be conducted in real time,The informal communication among planners, sales, production and customers is criticalto take informative and acceptable decisions.

Figure 5 Relationships between planning modules.

16062




Table 3

D&I Planning Capacity Planning MasterPlanning

Input Forecasts andorders

Long term:S&OP, aggregate

capacityparameters

Mediumterm: MPS,detailedcapacity

parameters

Aggregateplan (S&OP)and orders

Output Aggregate plan(S&OP), safety

stocks

Projectedcapacity plan

Projectedcapacity plan

MasterProductionSchedule

Feedbackloops

With Sales, withCapacity

planning andwith MasterPlanning

With D&IPlanning andmedium-term

capacity planning

With MasterPlanning andProduction(PP&S)

Witheverybody...

Granularity One year or morewith monthly /weekly buckets

One year or morewith monthly /weekly buckets

6 monthswith weekly /daily buckets

6 monthswith weekly /daily buckets

Frequency Monthly Monthly Weekly WeeklyTools Multi-echelon

inventory models,Statistical andqualitativeforecasting

Stochasticmodels, LP

models, MILPmodels

Stochasticmodels, LP

models, MILPmodels,heuristics

LP models,MILP models,heuristics

Simulation Appropriate at all levels in different ways

4.7 MP & ATPJosé Framinán, Ken Fordyce, Raphael Herding, Thomas Ponsignon, and Adar Kalir

License Creative Commons BY 3.0 Unported license© José Framinán, Ken Fordyce, Raphael Herding, Thomas Ponsignon, and Adar Kalir

The Central Planning Process (CPP) to manage the end to end demand supply network(DSN) for the production of semiconductor based packaged goods (SBPG) is focused onunderstanding and capturing exit demand, intelligently matching assets (WIP, inventory,and capacity across the network) to create a projected supply line linked to demand andsynchronizing (but not scheduling) activities across the DSN.

Figure 6 has a high level summary of the core components in CPP and their relationship.The “order handling activity” (where ATP occurs) captures the requests of products (orders)from customers. An incoming order is either accepted and given a commit date or identifiedas pending further review. Orders (accepted and pending) are sent to the “demand andinventory planning activity” where a comprehensive statement of exit demand is established.This demand statement is sent to the “master planning activity” which matches assets withdemand to create a projected supply line linked to demand. This is the basis for the commitdecision – that is what product, in what quantity, can be committed to the customer or madeavailable for the ‘automatic’ ATP process – that is when a commitment is made withoutgoing through the master planning process.

Although there are differences between firms, this is the basic process. Potential areas of




Figure 6 Summary of Master Planning and ATP flow.

improvement that will positively impact a firms’ responsiveness are:1. Improved flexibility in the projected supply used by the ATP, especially with regards to

level of personalization. For example, assume I make T shirts in three colors: red, blue,and white – where red and blue shirts are made from white shirts going through an endstep of dying and heating. I might project 100 reds, 100 blues, and 200 white as supply,but until the dye is done, this can be modified. Going a step back to material, untilthe shirt is cut, I have options to convert large to small, small to medium, etc. Oftenthis flexibility is only captured if the in the master planning or central planning engineactivity

2. On the order receiving side – being able to capture the “confidence”that the customerorder is firm and using that within planning process would be most beneficial – that ishow to incorporate uncertainty without creating confusion.

3. Faster and more intelligence central planning engines would be very beneficial – creatingtighter coupling.

Figure 7 provides a more detailed view of the key components in master planning and ATPto create a projected supply linked to demand and synchronize the activities of the demandsupply network (DSN), and planners workbench. From an operations research perspective,the most significant technical achievements occurred in the CPE (Denton et al., 2006 andDegbotse et al., 2013). Fordyce et al. (2011) has the best overall description of Order PlanningSystem (OPS). The business contribution of OPS is covered in Lyon et al. (2000), Dentonet al., (2006); and Fordyce (1998 and 2001). The work done by IBM in collaboration withArkieva on demand management captured a critical paradigm shift that successful demandmanagement was about a single integrated and flexible view of the key data sources (orders,forecast history, shipping, etc.), capturing sales estimates, collaboration, and developinginsight – where statistical forecasting methods was just one component (Fordyce and Sullivan,2016); this approach has carried forward to become best practice. Today the features andfunctions in OPS are standard best practice.

16062


Figure 7 Detailed Components and Flow for Master Planning and ATP.

4.8 MP & FactoryHubert Missbauer, Gerald Weigert, Jesus Jimenez, Lars Mönch, and Sebastian Knopp

License Creative Commons BY 3.0 Unported license© Hubert Missbauer, Gerald Weigert, Jesus Jimenez, Lars Mönch, and Sebastian Knopp

Interface to Master Planning (MP)

“Master Planning provides expectations or targets of what is required from each factory overtime.” (Master Planning – Breakout I). That is, quantities to be finished per product, period(usually weeks) and production unit (e.g., Fab, A/T, DC). Also the allocation to demandclasses (e.g., customer order driven vs. forecast driven) which might affect the priorities.

The subproblems in order to finally arrive at a satisfactory schedule are the following:Split up the MP quantities into production lots, defined in terms of product, lot size andrequired due date in days. Both backward scheduling and forward scheduling should bepossible.Order release (planning), where an order is a production lot. The extent to which releasesare planned for several periods depends on the applied method (short-term releasemechanisms vs. multi-period release planning). At this point the relationship betweenrequired output, target WIP and flow time norms must be considered. The underlyingmodel of the material flow through, e.g., the fab is crucial. Target flow times can befixed or load-dependent (see presentation Hubert Missbauer). Flow time anticipationusing flow factors assumes strong correlation between processing time and flow time ofan operation which is not always given. Order release usually is performed periodically.This can be complemented by event-driven releases (e.g., if otherwise machine idlenesswould occur), depending on the release mechanism.Due to the complexity of the material flow within a fab a global scheduling level (“LotPlanning”) is required – setting intermediate due dates for production phases (work




centers or groups of work centers). This requires an overview of the whole fab which thelocal dispatchers do not have. (See presentation Stéphane Dauzère-Pérès.) End-of-horizoneffects can occur, this requires look-ahead-feature.Based on the intermediate due dates for the released lots, detailed scheduling – dispatchingis performed locally per production phase. This is often rule-based and requires extensiveknowledge about the production process at this stage.Feedback (MES): The state of lots and work centers must be known at each time. Thisfeedback information is transferred to the respective decision levels. In any case this mustbe the release level since load-based release requires accurate state information. Whichdecision level should react to unplanned events (e.g., machine downtimes) is a difficultquestion; as a general rule it should be the lowest possible level. Minor disturbanceswill just affect the dispatching level whereas major machine downtimes might even affectcapacity and master planning.

Additional remarks

There are groups of wafers that share the first n operations; sometimes these base wafers arestored in order to benefit from variability pooling. This is an adaptation to individualizationof customer needs. We discussed if it makes sense to formalize this and to declare the basewafers as SKUs (which implies a second release decision at this point).

Lot-to-order matching is an important task. How to assign this task to planning levelsrequires further discussion.

4.9 Incorporation of Human Behavior in the Supply ChainHans Ehm, Cathal Heavey, Scott J. Mason, Oliver Rose, and Iris Lorscheid

License Creative Commons BY 3.0 Unported license© Hans Ehm, Cathal Heavey, Scott J. Mason, Oliver Rose, Iris Lorscheid

The aim of this seminar is to provide a reference model that describes the structure, processes,and basic skeleton of a supply chain in semiconductor industries. However, this referencemodel not includes a relevant complication, which follows from the uncertainty of humanbehavior within supply chains. Thus, complementing to the reference model, we aim forincorporating human behavior by exploring ways to identify relevant individual behavior andunderstand their effect on supply chain performance.

First, the identification of individual strategies clarifies the existing individual deviations.Also, we aim to reveal (informal) interactions that proofed to support the quality of de-cisions and were therefore implemented by individuals. A simulation model may analyzecircumstances that extend rational behavior, and aspects that may hinder rational behavior.Simulation experiments may quantify the risks for supply chain performance caused by lessrational behavior.

Overall, we aim for improving the circumstances to support good decisions of planners.Knowing about favorable individual behavior, we may relate this to the right level of educationthat is required for the decision makers. Results can define criteria for recruiting with regardto technical and soft skills, and identify training requirements. By learning about the effectof varying strategies, the awareness and sensitivity to acknowledge (un-)favorable aspects ofindividual behavior can improve. In the end, we aim to avoid negative human subjectivity.

16062




The first important step is to understand what humans actually do in our supply chainsto understand the relevance: Where is the “Ron” in the model? As we know, Ron may haveindividual interests, for which he under- and over-estimates, and thus potentially behavesrational in his own interest but not rational for the company. In particular within demand,capacity, or master planning we observe persons who may “pre-process” data by adjustingregulation screws that improves individual planning result, such as by adapting demand,capacity, yield, or – as common example – lead time. We see events where undesirablebehavior happens, and cases with undesired consequences. Also extreme cases such as hyper-optimistic or hyper-conservative planners are observable. Next to the planning scenario,other perspectives on the supply chain incorporate human behavior, such as negotiationsbetween sales persons and customers and their effect on (un-)successful pricing, and orderbehavior by customers resulting in varying demand patterns.

We see some challenges for incorporating human behavior. First, the design of incentivescan be relevant, so that individuals behave in the interest of the company rather thanchoosing fast and easy moves when fulfilling their tasks. Nevertheless, the design andsuccessful implementation of incentives is challenging. Next to unfavorable behavior, weknow about individual strategies that really stabilizes the supply chain. In compliance, forexample, individual adaptations are necessary to fulfill the compliance challenge on the onehand, but be right for the internal processes such as product planning on the other hand.Thus, we know that individual behavior may work in certain circumstances but may lead todamage in others. The question is: How many Rons (or “Hans”) do we need to keep thesystem running?

A further challenge with regard to collecting data for the identification of human behaviorcan be the identification of the “real” human factor that leads to the respective individualstrategy of interest. Dominant factors such as experience, age, or seniority may prevail otheraspects.

As concrete next step we propose to create a list of individuals that have an impact onthe supply chain, with a description of why and how they have an impact. The definition ofcases with positive and negative effects of human interactions on supply chain performancemay clarify the relevance of incorporating human behavior. We expect that positive casesof individual strategies will include situations comprising complications such as innovationsor complex tasks. Also, a literature review about similar studies about the effect of humanfactors on performance may support the research process.

4.10 From Reference Model to Systems (the SC System in 20201)Chen-Fu Chien, Can Sun, Peter Lendermann, Leon F. McGinnis, and Irfan Ovacik

License Creative Commons BY 3.0 Unported license© Chen-Fu Chien, Can Sun, Peter Lendermann, Leon F. McGinnis, and Irfan Ovacik

The Reference Model for the Semiconductor Supply Chain contains a set of objects and therelationships between those objects, along with behaviors and controls that describe howthe Semiconductor Supply Chain changes/evolves over time. These objects, behaviors andcontrols are described using a Domain Specific Language.

The Reference Model can then be used to instantiate/describe the physical supply chainof any semiconductor company. We refer to this as the Model Instance.

In academic settings and in practice, we usually talk about Math Programming, Queuing,




or Simulation Models. In practical settings, we also find these models embedded in planningapplications, wrapped with system and user interfaces. We will refer to these as AnalysisModels. Each Analysis Model consists of inputs, algorithms, and outputs and targets aspecific business decision.

Today, each Analysis Model potentially assumes a different Reference Model. For example,a JDA application may assume a different Reference Model than an SAP module even thougheach serves the same business function. Therefore, there is no guarantee that an AnalysisModel that works in one setting will work in a different setting without considerable effort.Similarly, there may be considerable different effort to implement a JDA application than anSAP module in the same company.

In the long term, once a Reference Model for the Semiconductor Supply Chain is in place,then each Analysis Model in the environment is expected to conform to the Reference Model.Each Analysis Model can use a subset of information available in the Reference Model oran abstraction of it. For example, a Production Planning and Scheduling model would onlyuse the relevant information for the Back End Supply Chain or a Master Planning modelwould treat each factory as an abstract black box, ignoring the details of each factory. Inboth cases, the Analysis Model would be consistent with the Reference Model. To the extentthat the Analysis Model conforms to the Reference Model, then the model would be reusablewhether it is interfacing with a Model Instance associated with company A or the ModelInstance associated with Company B.

4.11 Process Models Demand and Inventory Planning, CapacityPlanning, Master Planning

Ton de Kok, Israel Tirkel, Reha Uzsoy, and Stéphane Dauzère-Pérès

License Creative Commons BY 3.0 Unported license© Ton de Kok, Israel Tirkel, Reha Uzsoy, and Stéphane Dauzère-Pérès

The relations between demand planning, inventory planning, capacity planning andmaster planning

We considered the recurrent communication between Demand and Inventory Planning (DIP),Capacity Planning (CP) and Master Planning (MP) at monthly and weekly frequency.Starting with monthly regional sales forecasts, which are consolidated into product forecastsby a demand planner, DIP communicates monthly with CP to agree on a long-term capacityand sales plan. This process is iterative as capacity may not be able to accommodate theinitial sales plan. This results into updated regional sales plans which are input for MP. Werealized that our process is closely related to the TSMC process, which focuses on waferproduction. Companies with an integrated front-end and back-end start from customersales plans at SKU level, which is consolidated with respect to resource requirements andpossibly to SKU requirements in case multiple customers need the same IC. But similarly thelong-term consolidated sales plans must be translated into long-term capacity requirementsby capacity planning. A similar iterative process leads to updated customer (or regional)sales plans, which are input for MP.

MP processes the sales plans and derives time-phased key component requirements andkey resource requirements. The latter are discussed with CP to check if mid-term therequired resources are available. Again, an iterative process yields an updated MP, which is

16062




Figure 8 Process Modell Demand & Inventory Planning, Capacity Planning, Master Planning.

communicated with sales. This may lead to mid-term adjustments to the sales plans. Thisprocess repeats itself weekly.

We closed the loop by using the sales plans and masterplans of the current month asbenchmarks for the sales planning and capacity planning for the next month.

4.12 Process Models Demand and Inventory Planning, ATP, MasterPlanning

Kenneth Fordyce, Raphael Herding, Adar Kalir, and Thomas Ponsignon

License Creative Commons BY 3.0 Unported license© Kenneth Fordyce, Raphael Herding, Adar Kalir, and Thomas Ponsignon

Event-driven process chains Demand Planning – Master Planning – ATP Usage

New orders enter via the Order Handling System. A new order entry triggers the provision ofconfirmed and pending orders to Demand Planning. Demand Planning involves collaborativeprocess/human activities. It is typically run according to time-scheduled events. The outputfrom Demand Planning is a Demand Outlook, which is provided to Master Planning. MasterPlanning accepts further inputs from Capacity Planning and Production Management thatare not represented on this chart. Master Planning involves a solver. The output is aprojected supply or a supply menu (i.e., available supply with a set of conditions such aspricing and given lead times) depending on the business model of the company. The supplypicture is provided to the ATP Usage, which involves an Order Promising System. TheATP Usage activity is usually run according to an order fulfilment event. It results an orderresponse that is sent to the customer. Further information may be added to the process flow




Figure 9 Process Model Demand Planning – Master Planning – ATP Usage.

to facilitate the decision-making: confidence in orders may help to capture the uncertaintyof demand; price may support the prioritization of demand items in Master Planning; andFlexibility of projected supply may support the ATP Usage-related decisions. The processmodel is shown in Figure 9.

4.13 Process Models Master Planning and FactoryHubert Missbauer, Gerald Weigert, Jesus Jimenez, Lars Mönch, and Sebastian Knopp

License Creative Commons BY 3.0 Unported license© Hubert Missbauer, Gerald Weigert, Jesus Jimenez, Lars Mönch, and Sebastian Knopp

Event-driven process chains Master Planning – Factory

The master plan delivers quantities for the different facilities. Lots are generated based onthe master plan. The lots are equipped with due dates. Release dates for the lots are eithercomputed based on simple heuristics like backward and forward termination or based onoptimization approaches. Intermediate due dates are derived. The proposed process modelis shown in Figure 10.

4.14 Agents in the Level 3 Supply ChainIris Lorscheid, Scott Mason, Irfan Ovacik, and Can Sun

License Creative Commons BY 3.0 Unported license© Iris Lorscheid, Scott Mason, Irfan Ovacik, and Can Sun

The “level 3” supply chain is defined by Infineon Semiconductor as a semiconductor company’ssupply chain. Within this level 3 supply chain exists a number of potential types of agents

16062






Figure 10 Process Model Master Planning – Factory.

(e.g., capacity planners, inventory planners, master planners, demand planners, supply chainplanners, customer logistics managers, etc.). Regardless of its type, each agent has a set ofattributes that include (but are not limited to) memory (i.e., the agent’s knowledge base, ruleset, and facts), the agent’s role, and the coworkers with which the agent interacts. Further,each agent has abilities relating to individual strategies, input maintenance/analysis, andoutput analysis.

In terms of individual strategies, an agent can have specific strategy for dealing withinventory-related issues and ramp up/ramp down transitions, and another strategy foraccommodating customer prioritization contingencies. As an example, consider a masterplanning agent (Figure 11; other agents could be inserted into this figure without lossof generality). During the input maintenance and analysis task, the agent may a) querydatabases and other data sources, b) ask the sales and marketing department for a demandinput, and c) ask the factory planning department for a statement of available capacity.Once the master plan is complete, the output is analyzed for suitability. If the agent issatisfied, the results can be published to the production planning process (e.g., factory startsinformation) and/or the ATP process (e.g., a statement of supply). Otherwise, if the plan isnot deemed to be suitable, the agent returns to the input maintenance and analysis task.

4.15 Top Down Reference ModelHans Ehm, Cathal Heavey, Peter Lendermann, Leon McGinnis, and Oliver Rose

License Creative Commons BY 3.0 Unported license© Hans Ehm, Cathal Heavey, Peter Lendermann, Leon McGinnis, and Oliver Rose

A High Level Semiconductor Supply Chain Model exists at a semiconductor company whichcan be considered a common denominator among all the stakeholders in Semiconductor




Figure 11 Process Model Master Planning – Factory.

Supply Chain Management and an appropriate starting point for the development of aReference Model. However, business issues are typically identified on the lowest level processdescription level. A specific example would be an incident where it turned out that a changein shipment lot size for a particular product required different customs clearance procedurewhich was not taken into consideration during the Development and Master Planning stagefor that product.

To avoid such incidents, the team agreed that a Reference Model could help if it also goesto the lowest level of process detail. However, for a process alignment on this level a commonDSL (Domain Specific language) based on SYSML and containing PROS (Process, Role,Object, and State) would be needed. In a second stage this would also require an agreementbetween semiconductor manufacturers, software supplies, academia and other stakeholdersto take full benefit of such a DSL in semiconductor supply chain. This could turn into anuphill task for the following reasons which need to be carefully taken care of:

Low level process descriptions might contain some participants’ IP which owners wouldwant to protect. Once participants understand how a Top-Down Reference Model canbe developed without disclosing company-sensitive information this restriction can beovercome as nobody wants to be left behind.Development of a Reference Model would require considerable resources which are onlyrealistic if all participants can realise an ROI. After an initial publication and/or a paneldiscussion at WSC 2016, a European, US or Asia funded project could be the next step.The collaboration of International SEMATECH and JESSI, which ended up with theMIMAC models, could serve as a role model.

16062


Participants

Chen-Fu ChienNational Tsing Hua Univ., TW

Stéphane Dauzère-PérèsÉcole des Mines deSaint-Etienne, FR

Ton de KokTU Eindhoven, NL

Hans EhmInfineon Technologies –München, DE

Kenneth FordyceArkieva – Wilmington, US

José M. FraminánUniversity of Sevilla, ES

Cathal HeaveyUniversity of Limerick, IE

Raphael HerdingFernUniversität in Hagen, DE

Jesus JimenezTexas State University – SanMarcos, US

Adar KalirIntel Israel – Qiriat-Gat, IL

Sebastian KnoppÉcole des Mines deSaint-Etienne, FR

Peng-Chieh LeeNational Tsing Hua University –Hsinchu, TW

Peter LendermannD-SIMLAB – Singapore, SG

Iris LorscheidTU Hamburg-Harburg, DE

Scott J. MasonClemson University, US

Leon F. McGinnisGeorgia Inst. of Technology, US

Hubert MissbauerUniversität Innsbruck, AT

Lars MönchFernUniversität in Hagen, DE

Irfan OvacikIntel Corporation – Chandler, US

Thomas PonsignonInfineon Technologies –München, DE

Oliver RoseUniversität der Bundeswehr –München, DE

Can SunInfineon Technologies –München, DE

Israel TirkelBen Gurion University –Beer Sheva, IL

Reha UzsoyNorth Carolina State Univ., US

Gerald WeigertTU Dresden, DE

Jei-Zheng WuSoochow Univ. – Taiwan, TW

report from dagstuhl seminar 16062 modeling and analysis ...€¦ · report from dagstuhl seminar...

Documents